Identification of Cumulative Fruit Responses During the Storage Process Using Neural Networks

Abstract Neural networks are useful to identify complex nonlinear relationships between input and output of a system. Cumulative fruit responses such as water loss and ripeness during the storage process are characterized by nonlinearity. For identification, several patterns of these cumulative responses, as affected by environmental factors, are often obtained by repeating the same experiment several times under different environmental conditions. In this case, it is not well-known how many response patterns (training data sets) are necessary for acceptable identification. This paper explores an effective way for identifying such cumulative responses as the water loss and ripeness of tomato during the storage process using neural networks. First, the data for identification were obtained from a mathematical model and then the relationship between the number of response pattern and the estimation error was investigated. The estimated error becomes smaller when the number of response pattern is three or more. This suggests that three types of response patterns allow cumulative responses to be successfully identified. Besides, an addition of linear data (1,2, ., N) as input variable significantly improved the identification accuracy of such cumulative response. Finally, the identification of actual data were implemented based on these results and the satisfactory results were obtained.